Abstract
Research shows that verification of system outputs in human-AI interaction has long been framed as a trust-contingent behavior, where better-calibrated trust should reduce such verification. We tested this assumption through a mixed-methods survey of 153 frequent chatbot users. Contrary to the canonical prediction, we found no detectable association between trust and verification, with results robust across sensitivity analyses.
Key Findings
Three user practices—refinement, correction, and approval before automated actions—were widely endorsed and positively associated with satisfaction. The data revealed a substantive distinction between evaluative oversight (trust-decoupled, weakly tied to satisfaction) and interventionist oversight (weakly trust-correlated, strongly tied to satisfaction). A medium-to-large satisfaction-control gap indicated that effective task outcomes do not produce a felt sense of agency.
Qualitative Findings
Qualitative results identified instrumental mental models, failure-mode-specific doubt, and demand for epistemic infrastructure. We reframed user oversight as routine epistemic governance compatible with trust, deriving four design directions for scaffolded oversight in conversational AI.
Blogger's Review: This study challenges the traditional role of trust models in user verification behavior, emphasizing the proactive role of users in chatbot interactions. By redefining user oversight as epistemic governance, future designs can better meet user needs, enhancing satisfaction in human-AI interactions.